ChemSpider is an online database of over 20 million chemical structures assembled from almost a hundred data sources including chemical and screening library vendors, publicly accessible databases and resources, commercial databases and Open Access literature articles. Such a public resource provides a rich source of ligands for the purpose of virtual screening experiments. These can take many forms. This work will present results from two specific types of studies: 1) Quantitative Structure Activity Relationship (QSAR) based analyses and 2) In-silico docking into protein receptor sites. We will review results from the application of both approaches to a number of specific examples using the software outlined below. The QSAR analyses utilize the ChemModLab environment which is a free, web-based toolbox for fitting and assessing quantitative structure-activity relationships. Its elements include a cheminformatics front end to supply molecular descriptors, a set of statistical methods for fitting models, and methods for validating the resulting model. Five molecular descriptor sets are used with 16 math modeling methods to give a total of 80 QSAR models. The input is a file of compounds and a text file for biological activity. The in-silico docking experiments are conducted using a combination QSAR/Docking approach using the SimBioSys eHITS and Lasso software programs. The docking procedure allows for the screening of a complete molecular database to obtain the correct binding poses and estimated binding affinities. The ligand based screening tool utilizes a novel conformation independent 3D QSAR descriptor, ideally suited for scaffold hopping.

4. Example Search 1

5. Example Search 1

7. Example Search 2

8. Complex Search

9. Search Open Access Journals – ChemSpider

10. Search PubMed – ChemSpider

12. Quality is a Major Issue

14. Multi-level Curation and Approval

19. LASSO Descriptors on ChemSpider

21. LASSO Searching Method 2

26. General Trends?

27. ChemModLab + ChemSpider ECCR at North Carolina State Statistical Analysis Statistical Model Selected Compounds Initial Compounds Active Compounds Screen Compounds Collection

30. Results 2M commercial compounds taken from ChemSpider. Batches of 50k compounds processed. 42 batches on 12 processors takes about 12 minutes. Top 1000 predicted compounds selected Best model was Random Forest

31. PowerMV, 1000 predicted actives

32. Similarity Set 1

33. Similarity Set 2